Illustrative walkthrough

How I turn AI into portfolio-company margin

A hypothetical walkthrough of how I would approach a portfolio company, using Acme Technologies, a faux $40M-ARR SaaS whose gross margin is capped by cost-to-serve that grows with every new customer. Same method, any company.

Chad Newell · radiusos.ai/chad

The engagement

Diagnose

2 to 3 weeks

Find and rank the AI margin levers against your P&L.

Ship

90 days

Build one lever and measure it against a holdout, so the number is defensible.

Compound

ongoing

Sequence the remaining levers toward the number the exit needs.

Inside the 90-day sprint · Acme Technologies

The problem, on the P&L

Support$1.4M/yr · 14 reps · $13/ticket

Onboarding$720K/yr · 20 hrs/customer · 30-day time-to-value

Gross margin62%, target 70%+

Both cost lines scale with headcount. That is what caps margin.

The method

Instrument

Baseline the five metrics; build an eval harness on real resolved tickets. Their BI + eval layer.

Agent-assist first

AI copilot drafts replies for reps, grounded in Acme's docs. Humans stay in the loop. Helpdesk AI or Claude + retrieval.

Deflection, bounded

Self-serve resolution for high-volume, low-risk intents only. Confidence threshold + instant escalation. Billing and security always route to a person.

Onboarding automation

Automate config and data migration; AI drafts each setup plan from the customer's own data. Specialists move to expansion work.

The result

$650–740K/yr cost-to-serve removed

+1.5–2 ptsgross margin, most of it to EBITDA

30 → 12days to first value

At exit, that recurring margin is capitalized at your multiple.

$650–740KAnnual cost-to-serve removed

+1.5–2 ptsGross-margin gain

<$5K/moAI run-cost, engineered in

Several×Enterprise value at exit multiple

Why it works when others do not

Agent-assist before deflection

Prove quality with humans in the loop before anything reaches a customer.

Human-in-the-loop guardrails

The AI never answers billing, security, or contract questions. Under-promise, protect the brand.

Redeploy, don't cut

Freed capacity moves to expansion and retention, not layoffs. Better economics, a team that helps.

AI cost engineered in (97%)

I took my own product's AI cost down 97%. Your run-cost stays a fraction of the savings.